Predicting Rock Bursts

Sound sensors may be able to foretell mine explosions -- technology that could save lives and money

By Sui-Min Hsiung, PhD     image of PDF button

Dr. Sui-Min (Simon) Hsiung is a principal engineer in the Center for Nuclear Waste Regulatory Analyses at SwRI. He holds a doctorate in mining engineering from West Virginia University and bachelor's and master's degrees in mining engineering from National Cheng-Kung University, Taiwan. With more than 20 years of research and consulting experience in mining engineering and rock mechanics, he has authored more than 90 technical papers and reports.

Rock bursts -- sudden, explosion-like events that occur deep underground -- pose a serious safety hazard for mine workers and lead annually to millions of dollars in repair and cleanup costs for mine operators. They also can cause loss of production, premature mine closures and the abandonment of large mineral reserves. 

Characterized by the violent ejection of rock into underground openings from mine faces (walls) and pillars, rock bursts can occur in both hard rock and softer materials such as coal.

They are as sudden as they are powerful. For years, the mining industry has studied the mechanisms that cause a wall or pillar of solid rock to give way catastrophically, to detect and interpret the signals that indicate its approach and -- ultimately -- to predict its occurrence with confidence.

A research project, funded by Southwest Research Institute through its internal research and development program, has begun to develop a systematic methodology for rock burst prediction. The aim is to provide an early warning to mine management so that appropriate actions can be taken to reduce damage and save lives. 

Three conditions combine to trigger a rock burst: a high state of stress; high stiffness or strength of the rock formation; and the existence of free surfaces, such as those that surround rock pillars or rock faces where mining activities are carried out. 

During recent decades, extensive research has focused on predicting and controlling rock bursts. Significant advances have been reported in understanding the dynamic response of ground supports and in improving ground supports to contain or minimize rock burst damage, but little success has been seen in rock burst prediction. The event rate, signal characteristics and associated energy of acoustic emissions can herald incipient rock bursts, but past attempts to correlate rock burst activity to rising acoustic emission rates have seen limited success.

As mining goes deeper and mine structures become more complex (both of which produce higher stresses), rock bursts can be expected to become more severe and violent. Therefore, predicting a rock burst's timing and location becomes even more important so that personnel can be moved from the danger zone, ground supports can be reinforced and potential damage can be mitigated in advance of the actual event.

Rock failure

The SwRI team's approach focuses on monitoring and analyzing the clustering phenomenon of stress-induced fracturing in rocks and is based on the premise that the rock burst mechanism is the same as the failure mechanism of any rocks under stress; that is, the accumulation, clustering and coalescence of microfractures. 

Rocks are generally not homogeneous. They contain numerous voids, cracks and impurities. When rocks come under stress, localized failure most likely begins at existing crack tips or impurities because these are weaker than the rest of the rock structure or because they concentrate stresses. When these localized failures occur, sound, which is audible to sensors, is released. The pop that is emitted may be likened to the snap of a twig that has been bent beyond the breaking point. Therefore, formation of a microfracture is often called an acoustic event. 

As the stress in the rock continues to build, more areas of the rock are susceptible to localized failure; thus, more microfractures form. These microfractures begin to cluster and eventually coalesce to form major failure planes leading to macro-failure of the rock. At the time of macro-failure, rocks capable of storing large amounts of strain energy before failure could fail violently. Rocks that do not have the capability to store large strain energies, on the other hand, tend to fail nonviolently. 

The process of forming new microfractures, and the subsequent clustering and coalescing of these microfractures to develop failure planes, is called the microfracturing process. This process is similar in all types of rocks, whether or not they fail violently. Furthermore, the microfracturing process for rock samples tested in the laboratory is the same as for rocks underground. In other words, the microfracturing process is scale-independent, or fractal. The SwRI project takes advantage of this by analyzing the clustering phenomenon of the stress-induced microfracturing process in laboratory-scale rock specimens to identify its signatures and patterns. It then uses these patterns to determine the location and timing of an incipient failure. Tuff, a type of rock formed millions of years ago as volcanic ash compressed under its own weight, was selected for the test samples because it possesses high stiffness. When tuff samples fail, the failure is likely to be violent -- a form of rock burst.

Staff performed laboratory-scale compression testing by placing cylindrical rock specimens in a hydraulic press fitted with electronic sensors to record strain readings from gauges mounted on the samples. Downward pressure increased until the samples failed.

Temporal distribution of microfractures

Fractal patterns of the microfracturing events in both temporal (time) and spatial (distance) domains were assessed to identify precursors for rock burst prediction. In this study, 14 samples of tuff were tested under compression along a single axis. The uniaxial compressive stress was applied at a constant rate until the sample failed. The fractal pattern of the collected microfracturing events for each sample was analyzed in a progressive manner with time (and hence with increasing stress) for the duration of the test. 

Microfracturing events were recorded in some samples in the early stage of testing, when the applied stress was low. These events likely were related to the inherent weaknesses or flaws in the rock material. Once this early stage is completed, microfracturing activity diminishes as compression continues to build. This behavior is reflected in the variation of two important fractal characteristics (fractal dimension1 and prefactor1) with time. As shown in the accompanying graphic, fractal dimension is at high levels during the initial stages of loading. This indicates that microfracturing is taking place or that existing microfractures are closing under the compressive loads -- in a multitude of places throughout the sample, not just in a well-defined, 2-dimensional plane (which would be seen as a defined crack in the sample). Once the microfracturing activity stops, the fractal characteristics decrease to their lowest level and remain there for some time. 

Generally speaking, fractal dimension starts to increase at about 40 to 60 percent of the uniaxial compressive strength of the sample tested, indicating the formation of new microfractures. This increasing trend continues as applied stress increases. As the test progresses toward failure of the sample, the microfractures cluster and coalesce, producing 2-dimensional fracture planes. Formation of a fracture plane is accompanied by a decrease in either the fractal dimension (towards 2 dimensions)2 or the prefactor. Although a decreasing trend was observed for all but one sample, neither fractal characteristic consistently decreased in all cases, even with samples of the same rock type. Consequently, both fractal dimension and prefactor should be used together in a prediction model. 

The time delay for sample failure to take place, after a fractal characteristic reaches a peak value and starts to decrease, is a useful parameter. It may be used as a precursory indicator for failure, in other words, as a rock burst predictor. Sometimes the time delay between the decrease of a fractal characteristic and the actual rock failure is too long. This behavior may be related to the early formation of fracture planes that are not sufficient to coalesce into major fracture planes that can fail the sample; consequently, the microfracturing process continues. With the long delay, the associated fractal characteristic should not be used for forecasting rock bursts; in this regard, the time delay for the other fractal characteristic should be used as a precursory parameter instead. Ideally, both fractal characteristics should be used along with those for spatial distribution of microfractures.

Spatial distribution of microfractures

Before the spatial distribution of the microfractures can be analyzed, locations of the events must be determined. Two algorithms were examined for determining the locations of the events in this study. The first algorithm involves adopting a scheme that minimizes errors associated with the predicted source location of an event. This algorithm recognizes the uncertainties associated with measuring arrival times and the heterogeneity of the rock medium. It attempts to minimize the influence of random errors associated with these uncertainties. 

The second algorithm involves minimizing errors associated with predicted travel times from the event location to the sensors. This algorithm calls for inversion of the relative arrival time to determine the event location and the time of its initiation. Because the calculations of event locations and travel times are related, they must be estimated iteratively.

The accuracy of event location using either of the algorithms is affected strongly by difficulties in determining the velocity of wave propagation and arrival time. Because the material properties could be heterogeneous in a stressed geologic medium, the velocity could be different along different travel paths. To complicate the matter further, the velocity could change over time as microfractures form along the travel path. There is also an error in calculating the arrival time because the computer notes and records the arrival of a wave only after the amplitude exceeds some minimum value (a process that is often necessary to filter out noise).

Despite these difficulties, event locations are sufficiently accurate that the method is effective. Because the events are not located on a single (2-dimensional) plane2, the dimension is greater than 2. As the events start coalescing and forming the ultimate fracture plane, the associated fractal dimension should decrease and eventually equal 2 when the fracture plane is formed. The fractal dimension may not decrease all the way to 2, however, because more than one fracture plane typically develops during the failure process. The fractal dimensions for event locations decrease as fracture planes begin to form, and they drop to minimum values when the ultimate fracture plane is formed. After formation of the ultimate plane, existing fractures propagate and secondary fractures form, increasing the fractal dimensions. The decrease in fractal dimension toward 2 implies the formation of a major fracture plane that may grow to be the ultimate failure plane, which should aid in predicting rock failure. 

Predicting rock burst

Analyses of variations among fractal characteristics of microfracturing events in the temporal and spatial domains could provide a sound basis for rock burst prediction. Shortly before rock failure occurs, either fractal dimension or prefactor, or both, start to show a decreasing trend. This trend is a good indicator of rock instability; it can be used as a precursor for warning purposes. 

The fractal characteristics in the temporal domain and the fractal dimension in the spatial domain should be used together for predicting rock bursts. The event locations and their clustering, on the other hand, would provide ample evidence for the rock burst's location.

The next step is to test the prediction methodologies against microseismicity data gathered by the U.S. Bureau of Mines from past rock burst events. By comparing the laboratory-produced data to known events, researchers can assess how well and how far in advance the prediction data can be related to the event's timing. The Institute plans to begin this phase of the study during the next year.

Other applications 

This prediction methodology also may be applicable to engineering structures other than rock that have similar failure mechanisms based on clustering of microfractures -- for instance, reinforced concrete structures. Furthermore, this methodology may be used as an alternative for assessing structural damage, the extent of which is important to proper maintenance. Knowing the damage information may help in determining appropriate maintenance.

1 Fractal set can be defined according to a power law that expresses the number of events in terms of a constant (fractal intercept) divided by the dimension (such as time or distance) to some power. If the logarithm of the number of events is plotted against the logarithm of the dimension, the slope of the resulting curve is related to the "fractal" dimension and the intercept is called the "prefactor."

2A fractal dimension of 1 implies one dimension, such as is described by a line between two points; 2 implies two dimensions (described by a plane), and so on.

Published in the Spring 2001 issue of Technology Today®, published by Southwest Research Institute. For more information, contact Maria Stothoff.

Technics Spring 2001 Technology Today
SwRI Publications SwRI Home